Brief Review — Classifying Heart Sound Recordings using Deep Convolutional Neural Networks and Mel-Frequency Cepstral Coefficients
MFCC+CNN
Classifying Heart Sound Recordings using Deep Convolutional Neural Networks and Mel-Frequency Cepstral Coefficients
MFCC+CNN CinC’16, by Palo Alto Research Center
2016 CinC, Over 150 Citations (Sik-Ho Tsang @ Medium)Heart Sound Classification
2013 … 2023 [2LSTM+3FC, 3CONV+2FC] [NRC-Net] [Log-MelSpectrum+Modified VGGNet] [CNN+BiGRU] [CWT+MFCC+DWT+CNN+MLP]
==== My Other Paper Readings Are Also Over Here ====
- Heart sounds is converted as MFCC and input to CNN for classification.
Outline
- MFCC+CNN
- Results
1. MFCC+CNN
1.1. Segmentation
Each PCG waveform is firstly segmented into the fundamental heart sounds (S1, Systole, S2 and Diastole) using Springer’s segmentation algorithm [2].
- Segmentation was used to ensure that each 3-second heart sound segment began at S1.
1.2. MFCC
- 13 MFCC feature values are extracted for each sliding window.
In total each heat map consists of 300 time frames represented on the x-axis, and 13 MFCC filterbanks represented on the y-axis.
1.3. CNN
- A single channel 6×300 MFCC heat map is used as input and a binary classification is the output.
A standard architecture is used consisting of two convolutional layers, each followed by a max-pooling layer, followed by two fully connected layers before final classification.
- (Please read the paper for the detailed model architecture.)
2. PhysioNet Results
Results for the proposed top scoring submissions made to the PhysioNet challenge server for both Phase I and Phase II are depicted in Table 2.
